SIAgentNews.com
← All news

DeepSeek founder launches open-source personal AI agent Linger

Former DeepSeek project lead Tony Zhang launched Linger to bring local AI agents to personal computers, aiming to bypass cloud costs.

· First reported by 36Kr

Papercut shadow box: picture for the story 'DeepSeek founder launches open-source personal AI agent Linger'
AI-made papercut shadow box for this story. No real people or products are shown.
$6 millionseed funding for Simplex AI

The short version

  • Linger is an open-source personal AI agent that can operate software, browsers, and manage emails on a user's local computer.
  • The product features a local inference engine called PolyStrata, which allows large models to run on standard hardware.
  • Simplex AI secured $6 million in seed funding led by GL Ventures in January 2026.
  • Tony Zhang left DeepSeek in March 2024, before the release of the high-profile DeepSeek-R1 model.

Why it matters

This move signals a shift from competing on massive foundation models to creating practical, local applications. For users, it offers a way to use powerful AI assistants without recurring per-token costs from cloud providers.

From DeepSeek to Simplex AI

Tony Zhang, a former project lead at DeepSeek, has officially unveiled his new venture, Simplex AI. Zhang was a key figure in developing DeepSeek-VL, a vision-language model designed to understand images and charts alongside text. He left the company in March 2024, nearly a year before the release of the viral DeepSeek-R1 model.

After departing DeepSeek, Zhang returned to Zhejiang University to teach and pursue independent research. He eventually founded Hangzhou Topology Lingxi Technology Co., Ltd. in February 2025 to focus on AI agent research. His new company has already secured $6 million in seed funding led by GL Ventures to support its development.

While many of his former colleagues at DeepSeek have remained with the company or joined other tech giants, Zhang is the only departed core author to publicly launch a startup. He has stated that his experience at DeepSeek taught him the importance of focusing on doing one thing well. He now applies this philosophy to creating high-quality research and practical AI products.

How Linger works locally

The centerpiece of Zhang's new venture is Linger, an open-source personal AI agent. Unlike many AI tools that rely on expensive cloud infrastructure, Linger is designed to run on a user's personal computer. It can operate software, read authorized emails, and organize information into a personal knowledge base.

To make this possible on standard hardware, the team developed a local inference engine called PolyStrata. This engine was built by an 18-year-old developer on the team who extended an existing open-source engine. PolyStrata works by compressing models and optimizing how they are stored in a computer's memory.

For example, the engine can handle Mixture of Experts (MoE) models by keeping some modules in the graphics card's video memory while storing others on the hard disk. This allows a computer with an RTX 3060 graphics card to run a 125-billion-parameter model at a speed of 45.5 Tokens per second. This setup helps users avoid paying per-token fees to cloud providers.

The 'Another Me' assistant

Linger features a primary assistant called Another Me, which functions as a digital double for the user. It can proactively identify tasks, such as finding an upcoming deadline in an email and sending a reminder. If a user repeatedly ignores a specific type of reminder, the system uses a mechanism called "Dreaming" to adjust its behavior based on that feedback.

The system is also designed to handle complex workflows by coordinating multiple AI assistants simultaneously. These different agents can be assigned specific roles, such as coding or checking the results of a task. This collaborative approach allows the software to manage multi-step projects that a single AI might struggle to complete alone.

While the product is currently open-source, Zhang has noted that the team plans to support mobile phone connectivity within a month. He also addressed early installation bugs on social media, stating that the team has reproduced the issues and is working on fixes. The goal is to provide a persistent assistant that keeps working without constant human intervention.

Market shift and competition

Zhang's move reflects a broader trend in the AI industry where the focus is shifting from foundation models to applications. While the previous few years were defined by a race for computing power and model size, many new teams are now using open-source models to solve specific problems. Simplex AI is following this path by building tools that reduce costs for the end user.

However, the personal AI assistant space is becoming increasingly crowded with existing competitors. Other products already offer the ability to perform tasks on behalf of users while supporting local models. Some of these competitors are already established in the Chinese market, creating a competitive environment for Zhang's new product.

Simplex AI plans to offer a hosted version of Linger in the future, though pricing has not yet been announced. For now, the company allows users to deploy the software locally for free. Zhang intends to balance this by restricting other enterprises from using the source code to provide commercial hosting services without authorization.

The path forward for local AI

The success of Linger depends on whether ordinary users find the local setup practical enough to replace cloud-based tools. While it eliminates per-token fees, users still have to manage their own hardware costs and electricity usage. Additionally, any data sent to third-party applications or external models will still leave the local device.

Zhang's goal is to move away from a model where most AI power is concentrated in data centers. By bringing AI back to personal computers, he hopes to give users more autonomy over their data and their costs. This approach aims to make AI a constant companion that lives on the user's own hardware rather than a service they rent.

The industry will be watching to see if PolyStrata can successfully optimize enough models to make local agents a standard experience. As open-source models continue to improve, the barrier to entry for creating high-quality local agents will continue to drop. Zhang's journey from a DeepSeek core member to a local AI entrepreneur highlights the growing diversity of the AI ecosystem.

Try it

Check out the PolyStrata engine on GitHub to see how it handles model compression for different hardware configurations.

Watch

Will the 18-year-old developer's PolyStrata engine become the standard for local inference?

How will Simplex AI price its upcoming hosted version of Linger?

Will other DeepSeek core members follow Zhang's path into independent product startups?

Sources and links

Read the original at 36Kr

First reported by 36Kr. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news